|Publication number||US7853835 B2|
|Application number||US 12/194,642|
|Publication date||Dec 14, 2010|
|Filing date||Aug 20, 2008|
|Priority date||Jun 29, 2004|
|Also published as||US7418627, US20050289390, US20090089609|
|Publication number||12194642, 194642, US 7853835 B2, US 7853835B2, US-B2-7853835, US7853835 B2, US7853835B2|
|Original Assignee||Hitachi, Ltd.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (17), Non-Patent Citations (1), Referenced by (2), Classifications (14), Legal Events (1)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This is a continuation of U.S. application Ser. No. 11/065,352, filed Feb. 25, 2005 now U.S. Pat. No. 7,418,627. This application relates to and claims priority from Japanese Patent Application No. 2004-190633, filed on Jun. 29, 2004. The entirety of the contents and subject matter of all of the above is incorporated herein by reference.
The present invention relates in general to a computer system having a malfunction tolerance capability in an application system; and, more particularly, the present invention relates to a computer system with a program having a failover function, so that in case a malfunction occurs in the program of a computer or an operating system which is running an application, another computer is allowed to takeover the application being run.
A computer system which requires high reliability includes a currently-active system computer for executing a process (application) and a standby system computer which is capable of taking over the processing in case a malfunction occurs in the currently-active system. A procedure which is executed from the time of detection of a malfunction occurring in the currently-active system to the time the standby system is caused to take over the processing is provided by a cluster program. When the application uses data on a disk, the disk is shared between the currently-active system and the standby system. In order for the standby system to take over the processing in case a malfunction occurs in the currently-active system, it is necessary to select a standby system from the cluster computers, and, with respect to of resources used by the application and the operating system (OS), to take over the resource, which cannot be used at the same time (shared resource), such as a shared disk and an IP address. In order to realize higher reliability, it is also necessary to ensure that the currently-active system and the standby system do not use the shared resource at the same time, in the case where a malfunction occurs interrupting a path on which the standby system monitors a malfunction of the currently-active system (network split).
A method of selecting a standby system which takes over a process by exclusively taking over a shared disk in a cluster is performed by many cluster programs. As examples, reference is made to the below-listed patent document 1 and non patent document 1.
Patent document 1 describes a technique in which a mechanism for stopping a currently-active system, from a standby system, is used so that the standby system resets the currently-active system for releasing a shared resource owned by the currently-active system, and then the standby system owns the released shared resource for exclusively controlling the shared resource.
The non-patent document 1 describes a technique in which, in the case where a malfunction occurs in a currently-active system so as to perform a failover, a cluster program uses the commands RESERVE and RESET of available SCSI commands to exclusively control an access right to a shared disk. Here, RESERVE is a command for reserving an access right to a disk so that a RESERVE disk reserved by a certain computer denies an access and RESERVE from another computer. RESET is a command for releasing an access right of a disk so as to release an access right of the RESERVE disk.
[Patent document 1 [U.S. Pat. No. 6,138,248
[Non-patent document] Microsoft, Support Technical Information, 309186 (online, http://support.microsoft.com/kb/309186/en-us)
In patent document 1, in a cluster computer system, in a case where the standby system cannot monitor the currently-active system, it stops the currently-active system to obtain exclusive control of the shared resource. In a case network split occurs in a cluster of two computers, which constitute each other's standby systems, each of the systems resets the other, so that all of the systems can be reset. The process will be suspended at the time of a network split, so that high availability can not be reached.
Although the standby system resets the currently-active system, the currently-active system will not reset the standby system. When considering a cluster of a currently-active system and two standby systems capable of taking over it (standby systems 1 and 2), in the case of splitting a cluster of two computers of the currently-active system and the standby system 1 from the standby system 2 due to a network split, the standby system 2 resets the currently-active system to perform a failover. When the currently-active system is reset by the standby system 2, the standby system 1 also detects a malfunction of the currently-active system to perform failover. As a result, the standby systems 1 and 2 become currently-active systems at the same time, so as to cause a double access to the shared resource.
In accordance with non-patent document 1, in a cluster computer system, a standby system which cannot monitor the currently-active system includes a process forcefully releasing the control right of the currently-active system to a shared disk by use of the command RESET of the SCSI commands and a process of obtaining the control right of the shared disk released by issuing the RESERVE command of the SCSI commands from an arbitrary standby system. A system which takes over the shared disk, that is, a system which takes over the processing is determined by the two processes. When the latter RESERVE process is invalidated by the former RESET process, excessive failover occurs in such a manner that the process in which a take over is once performed with respect to a certain standby system by the command RESERVE is re-taken over by another standby system. To prevent this, enough time from the former RESET process to the latter RESERVE process is necessary to ensure that all of the standby systems complete the issuance of the RESET command. Irrespective of whether a network split actually occurs, the failover time can be delayed for a fixed time.
In accordance with this method, in a case network split occurs, failover can be performed. A further process for taking over succeeding the shared resource, other than a shared disk, e.g., of taking over an IP address, is necessary. However, the time required for completion of failover is increased so as to delay the failover time.
The present invention provides a high availability computer system, including an active system (active computer) and standby systems (standby computers) having a function for providing a quick failover. The active/standby computers share at least one resource, and they are combined with a path heartbeat for mutually monitoring each other and with a reset path for mutually stopping computer operations. For example, the shared resource may be a disk drive.
According to a representative embodiment of the present invention, each computer in the high availability computer system is provided with an index (reset priority) for determining the order in which a reset command is issued to other computers. The reset priority has a value uniquely determined in the high availability computer system. For example, the priority may be determined in the order of the IP addresses of the computers. Each computer in a cluster, other than the active computer, sets the delay time for reset issuance based on its own reset priority, in case the computer detects a malfunction in the active computer, based on its detection through the heartbeat path that the heartbeat of the active computer is lost.
A computer whose timer detects the elapse of the set reset delay time issues a reset to the active computer (malfunctioning system) in which a malfunction occurs. The reset delay time set to each computer has a time difference from the reset delay time to be set to the other computers so that more of the computers in the cluster can perform a reset at the same time. Preferably, the time difference between any two computers having an adjacent priority is a fixed time difference that is sufficient for determining whether the reset performed by the prior computer is has successful or failed.
When a certain computer resets a malfunctioning computer, the operation of the malfunctioning computer is stopped to end the use of a shared resource. The process of stopping the operation may be performed by turning off the power or by shut down of the OS.
The computer which has issued a reset (resetting system) communicates the resetting of the malfunctioning system to the other computers. All of the computers which have received a reset notification with regard to the malfunctioning system stop the reset timer before issuing a reset to the malfunctioning computer. This prevents a reset from being performed plural times for the same machine.
After the malfunctioning computer is reset, the standby computer which takes over the processing of the malfunctioning computer takes over the shared resource and the processing which has been performed in the malfunctioning computer.
The standby computer which takes over the processing may be the resetting computer which has reset the malfunctioning system or it may be the other computer identified by the resetting computer.
Thus, a high availability cluster computer system is provided in which the computer having the highest reset priority among normally operating computers inevitably resets a malfunctioning system, and one standby computer including a resetting computer which is able to take over the processing of the malfunctioning computer exists, so that failover to the standby system is performed for taking over the processing.
It is to be understood that the drawings and description related to the present invention have been simplified to show suitable elements for providing a clear understanding of the present invention, and that certain known elements are omitted which do not embody the present invention. This technique includes some conventional elements which should be changed to others which seem to be desirable and/or necessary for implementing the present invention. Those elements, which are known and which cannot facilitate a understanding of the present invention, will not be described here. The accompanying drawings will be described below in detail.
The NIC 1103 is used to enable the application 1102 to communicate with the outside. The NIC 1104 is used for communication to enable the cluster program 1110 to monitor other computers. The NIC 1105 is used to enable the cluster program 1110 to reset other computers. The NICs 1103 and 1104 are shown as separate elements for facilitating an understanding of the subject matter, but may be the same element.
The reset part 1106 has a function of receiving a reset commend from the cluster programs 1210 and 1310 of the other systems to stop the system A. The function of stopping the system A is realized by application of a forceful stopping to the OS 1107.
The cluster program 1110 has three modules. The cluster program will be described later with reference to
The failover part 1113 further has a cluster status management table 1114 for managing the status of a cluster.
The delay time (reset delay time) from detection of a malfunctioning system to issuance of a reset instruction from the main system to the malfunctioning system is stored in the reset timer 25. In case a malfunction occurs so that heartbeat detection of the system C is impossible, there is a difference in the reset delay time set to the timers of the systems. The reset delay time in the systems is stored so that the issuance of a reset instruction triggered by plural systems is effected with a time interval. In the present embodiment, specifically, the reset delay time of the systems is stored so that reset is performed in a certain order based on the reset priority 23.
A system with a priority one rank above that of the main system instructs a reset. Over a fixed time difference to ensure that reset is performed, it is possible to ensure that reset is performed in accordance with the reset priority 23. The reset delay time set to the timer of the systems may be set so as to establish a time difference for the systems along system hardware according to the reset priority set to the systems. Reset interval definitions 1116, 1216, and 1316 hold information on the time difference. Alternatively, the reset delay time of the systems may be directly set to the reset interval definitions 1116, 1216, and 1316 by the user.
The reset priority 23 may use a value which is not doubled and is uniquely determined in all of the cluster systems. The reset priority 23, which is statically provided by the user, is provided by the priority definition 1115 to the cluster program 1110.
The reset priority 52 may have the same value as that of the reset priority identifier 23 and may have a value uniquely determining the reset priority 23 using the reset priority 52. When using an IP address as the reset priority 52, the difference relation is used to uniquely determine the reset priority 23. The user statically provides the reset priority 23 based on the priority definition 1115. A method of dynamically determining the reset priority 23 by the cluster program may also be used. In this case, information included in the priority definition 1115 may be used.
System addition and deletion of information in the status management table 1114 are performed as follows. The cluster program 1110 starts monitoring with the cluster program of a new system using the monitoring part 1111 to add the new system to the table. When the cluster program 1110 resets a malfunctioning system using the failover part 1113, or when the cluster programs of other systems notify the cluster program 1110 that the malfunctioning system is reset, information relating to the malfunctioning system is deleted.
For simplifying the description, this embodiment illustrates an example in which the failover part 1113 has one cluster status management table 1114. However, the information 21 to 25 included in the cluster status management table may be divided into plural tables for management and may be in a cluster program other than the failover part.
When executing the cluster program 1110, malfunctions of each other's systems are monitored. The monitoring part 1111 periodically communicates with the cluster programs 1210 and 1310 of the other systems via the communication part 1112 to perform system monitoring of the other systems. The status of each of the systems obtained in step 31 is registered into the system status 22 corresponding to the system identifier 21 in the status management table 1114 (step 301)
The system status 22 of each of the systems obtained in the step 301 is referred to so as to judge whether a system in which a malfunction occurs (malfunction system) exists or not (step 302). When a malfunctioning system does not exist, it is determined that all systems are being normally operated. The routine is returned to the step 301 to periodically continue monitoring. When a malfunctioning system exists, the monitoring part 1111 calls the failover part 1113 (the dotted line in the drawing) to execute step 401 of the failover processing (step 303)
The monitoring part 1111 executes the step 303 to return to the step 301 for monitoring the status of the other systems in the cluster again.
The instruction to the failover part, which has been performed in step the 303, is judged in step 401. In the step 401, whether a new malfunctioning system has been detected or not is judged. When a new malfunctioning system exists, the failover part 1113 refers to the reset priority 23 of the management table 1114 (step 402). The time during which the main system resets the malfunctioning system is set to the reset timer 25 of the cluster status management table 1114 based on the reset priority (step 403) so as to perform step 404. In the step 401, when a new malfunctioning system does not exist, the step 404 is performed.
In the step 404, whether the time of the reset timer set in the step 403 has elapsed or not is judged. At the time to issue a reset, a reset instruction is sent via the communication part 1112 from the status management table 1114 to the reset parts 1206 and 1306 of the other systems indicated by the reset side identifier 24 of the system identifier 21 to be reset (step 405).
The reset parts 1206 and 1306, which have received the reset instruction, stop the operation of the main system so as to stop the use of the shared resource. To stop the operation of the main system, the power may be turned off, the software may be reset, the OS may be shut down, or the OS may hang up.
In the step 405, after the reset is successful, the failover part 1113 sends a notification via the communication part 1112 to the cluster programs of other systems that the reset of the malfunctioning system has been performed (step 406). The reset timer 25 of the reset malfunctioning system in the status management table 1114 is cleared (step 407), and the processing returns to the step 401.
In the step 404, when the time of the reset timer set in the step 403 has not elapsed, whether reset completion is performed from other systems or not is judged (step 408). When other systems having a reset priority higher than that of the main system exist, the cluster programs of the other systems precedently execute steps 404 to 407. The malfunctioning system already may have been reset.
When the cluster programs of the other systems indicate a completion of the reset, the reset timer of the reset malfunctioning system is cleared (step 407) so as to not reset the malfunctioning system again, and the processing will return to the step 401.
In step 408, when the cluster programs of the other systems have not indicated reset completion, the malfunctioning system has not been reset, and the processing will to return to the step 401 without performing any process.
For simplifying the description,
When a network split occurs at time T0, times T1, T2, and T3 are set to the reset timers of the systems A, B, and C, respectively, based on the reset priority. When the systems are normal, reset is performed to a malfunctioning system, as seen from its own computer at the set time.
When the systems B and C malfunction, as seen from the system A, the system A resets them at the time T1 (arrows 601 and 602). In case malfunctions which cannot be reset by the system A occur at the same time, the system B performs a reset at the time T2 (arrows 611 and 612). In case a malfunction which cannot be reset by the system B occurs, the system C performs reset at the time T3 (arrows 621 and 622). The system A owns the shared resource at the time T0 to T2. The system B owns the shared resource at the time T2 to T3. The system C owns the shared resource at the time T3 to T4. Failover is thus performed. After the time T4, the reset path cannot be normal, which is not targeted in the present invention. According to the above-described embodiment of the present invention, the following effects are provided.
In a high availability cluster computer system having a reset path so that, in case the heartbeat for the system monitoring is lost, the malfunctioning system is reset in accordance with the reset priority. This ensures that only systems in a cluster, in which the system which has performed a reset, use the shared resource. Failover can be realized at the time of a network split.
The user statically defines the priority definition to the cluster program. It is possible to realize failover to freely set the reset priority to control the side subject to failover.
The user defines a reset interval to the cluster program to control reset the timing.
According to the present invention, when an active system computer cannot be monitored, failover can be realized to a cluster in which a system having high reset priority exists. Quick failover is thus possible. It is expected that the present invention can be widely embodied as a high availability computer system.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US4639859||May 24, 1984||Jan 27, 1987||Rca Corporation||Priority arbitration logic for a multi-master bus system|
|US4817091||May 19, 1987||Mar 28, 1989||Tandem Computers Incorporated||Fault-tolerant multiprocessor system|
|US5583987||Dec 21, 1994||Dec 10, 1996||Mitsubishi Denki Kabushiki Kaisha||Method and apparatus for initializing a multiprocessor system while resetting defective CPU's detected during operation thereof|
|US6138248||Dec 30, 1997||Oct 24, 2000||Hitachi, Ltd.||Common disk unit multi-computer system|
|US6370656 *||Nov 19, 1998||Apr 9, 2002||Compaq Information Technologies, Group L. P.||Computer system with adaptive heartbeat|
|US6393590 *||Dec 22, 1998||May 21, 2002||Nortel Networks Limited||Method and apparatus for ensuring proper functionality of a shared memory, multiprocessor system|
|US6550017 *||Jun 29, 1999||Apr 15, 2003||Sun Microsystems, Inc.||System and method of monitoring a distributed fault tolerant computer system|
|US6584587 *||Oct 14, 1999||Jun 24, 2003||Sony Corporation||Watchdog method and apparatus|
|US6738923 *||Sep 7, 2000||May 18, 2004||International Business Machines Corporation||Network station adjustable fail-over time intervals for booting to backup servers when transport service is not available|
|US6920580 *||Aug 20, 2001||Jul 19, 2005||Network Appliance, Inc.||Negotiated graceful takeover in a node cluster|
|US6922791 *||Aug 9, 2001||Jul 26, 2005||Dell Products L.P.||Failover system and method for cluster environment|
|US6959344||Oct 28, 2002||Oct 25, 2005||Hitachi, Ltd.||Method and apparatus for storage system|
|US7134054 *||Jan 13, 2004||Nov 7, 2006||Hitachi, Ltd.||Method for monitoring fault of operating system and application program|
|US20030159084 *||Jan 10, 2000||Aug 21, 2003||Sun Microsystems, Inc.||Controlled take over of services by remaining nodes of clustered computing system|
|US20030212923||May 8, 2002||Nov 13, 2003||Dell Products L.P.||Method of thermal shut down and failure status capture in a storage enclosure|
|US20040103345||Nov 21, 2002||May 27, 2004||Dunstan Robert A.||Method, apparatus and system for ensuring reliable power down of a personal computer|
|US20050010838 *||Apr 23, 2004||Jan 13, 2005||Dot Hill Systems Corporation||Apparatus and method for deterministically performing active-active failover of redundant servers in response to a heartbeat link failure|
|1||"How Cluster Reserves A Disk and Brings A Disk Online", previously published under Q309186, pp. 1-3, printed Jan. 18, 2005.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US8176302 *||Sep 26, 2006||May 8, 2012||Nxp B.V.||Data processing arrangement comprising a reset facility|
|US20090019274 *||Sep 26, 2006||Jan 15, 2009||Nxp B.V.||Data processing arrangement comprising a reset facility|
|U.S. Classification||714/55, 714/12, 714/11, 714/56, 714/13, 714/4.1|
|International Classification||G06F11/07, G06F11/00|
|Cooperative Classification||H04L69/40, G06F11/2046, G06F11/2028|
|European Classification||G06F11/20P2E, G06F11/20P12, H04L29/14|